Un outil d'IA "surhumaine" détecte une maladie cardiaque en moins de 2 secondes

Artificial intelligence may soon be able to diagnose heart disease, high blood pressure, and diabetes through a simple five-second scan of a person's face or a routine electrocardiogram (ECG). New research presented at the European Society of Cardiology (ESC) congress in Munich suggests that AI tools can extract critical health insights from medical data much faster and more accurately than human doctors. The findings, revealed during the world's largest heart conference, highlight the transformative potential of AI in modern medicine. One AI system, currently undergoing trials in the National Health Service (NHS), claims to identify individuals at risk of heart failure and heart valve disease from an ECG scan. Another AI model demonstrated the ability to detect high blood pressure and type 2 diabetes by analyzing subtle changes in skin color and blood flow visible in a brief facial video. In the United Kingdom, approximately 16 million people suffer from hypertension, four million have type 2 diabetes, and over eight million live with heart disease, many of whom remain undiagnosed. The current diagnostic process for heart-related issues often involves long waiting times, with hundreds of thousands of patients on NHS cardiology waiting lists. This delay increases the risk of complications such as strokes and heart attacks. The AI program under trial in London and Bristol uses data from 10.6 million ECG scans and has shown promising accuracy, identifying 81% of patients with heart failure and 90% with heart valve disease. If successful, the technology could be integrated into the NHS within five years, allowing for rapid identification and subsequent prioritization of patients requiring further testing. Traditional ECG machines measure the electrical activity of the heart, detecting irregularities in rhythm and rate. However, they have historically failed to identify structural heart problems such as heart failure or valve disease, which require more detailed imaging via echocardiograms. These scans are often delayed due to resource constraints, leaving many patients vulnerable to worsening conditions. Dr. Ahmed El-Medany, who led the ECG analysis at Imperial College London, emphasized the significance of the AI's performance. He noted that the AI's ability to recognize patterns beyond human perception could revolutionize diagnostics. The next step is to evaluate how effectively the AI functions within the NHS environment. Portable ECG devices are also being tested to determine if they can provide sufficient data for AI systems to detect heart disease in home settings. Dr. Roy Jogiya, chief medical researcher at Heart Research UK, described the implications of AI screening as potentially transformative. He pointed out that while ECGs are among the most accessible medical tests, AI is revealing untapped diagnostic value. Dr. Sonya Babu-Narayan, clinical director at the British Heart Foundation, expressed enthusiasm about the AI's ability to produce results almost instantaneously. She stressed the importance of early detection in improving outcomes for patients with heart conditions. While the AI cannot replace definitive diagnoses, it offers a powerful tool for identifying high-risk individuals who can then receive timely follow-up care. At the ESC congress, researchers also shared preliminary findings on another AI model capable of diagnosing high blood pressure solely from facial video footage. By analyzing microvascular responses and skin tone variations, the AI can infer cardiovascular status without the need for conventional blood pressure measurements. This dual approach, leveraging both ECG data and visual cues, could significantly reduce diagnostic delays and improve access to care. As the technology continues to evolve, experts anticipate broader applications, including opportunistic screenings during routine check-ups or emergency assessments. The integration of AI into mainstream healthcare faces challenges, including ensuring accuracy, addressing ethical concerns, and adapting existing infrastructure. Nevertheless, the initial success of these AI models underscores their potential to enhance diagnostic efficiency and patient outcomes. Researchers plan to refine the algorithms further and conduct larger-scale trials to validate their effectiveness across diverse populations. They also aim to develop user-friendly interfaces that allow healthcare providers to easily incorporate AI-assisted diagnostics into daily practice. As the NHS and other healthcare systems explore ways to alleviate strain on resources, innovations like AI-driven diagnostics offer a glimpse into a future where complex health assessments can be conducted swiftly and efficiently. The ongoing refinement of these technologies promises to reshape the landscape of preventive medicine and chronic disease management.